Instructor

Ankur Kulkarni

Semester

Autumn ‘20

Course Difficulty

Quite easy, at least for someone with a reasonable background in linear algebra and probability.

Time Commitment Required

Not more than 1 hour per week, in addition to lectures

Grading Policy and Statistics

Since a vast majority of the students in the class were CSE thirdies and fourthies, AAs and ABs were quite challenging to obtain.
5 AAs, 12 ABs and 25 BBs were awarded in a class of 57 students. No failure grades were awarded.

Attendance Policy

None

Pre-requisites

Sound knowledge of Linear Algebra (MA106, SC639 etc.) and Probability (EE223 or equivalent) is a must. At the start, the professor demanded that students should have done a course in optimization as well, but this requirement was modified into only students of third year and above.

Evaluation Scheme

We had four take-home objective quizzes on Microsoft Forms. As is standard with Prof. Kulkarni’s courses, he also expects each lecture to be scribed (Beamer slides) by a group of up to 3 students. There was also a paper review and presentation, again in groups of up to 3. Weightages of individual components were not disclosed.

Topics Covered in the Course

Basics of static games: Zero-sum and non-zero sum games, concept of Nash equilibrium and Stackelberg equilibrium.
Multi-act games: extensive form of games and information sets. Aumann’s common knowledge, rationality, bounded rationality.
Dynamic games: Incomplete information, Bayesian Nash equilibrium. General formulation of dynamic games: sub-game perfectness, open-loop, closed-loop and feedback Nash equilibria, informational properties of Nash equilibria, informational non-uniqueness.
Information structures: static and dynamic information structures.
Dynamic stochastic team problems: introduction, person-by-person optimality, Witsenhausen problem, signalling, connections to economics and information theory.

Teaching Style

Lectures were conducted twice a week on Microsoft Teams. The professor would write on a virtual whiteboard and present his screen, and later share these hand-written notes. Teaching was very good and lectures were interesting, most certainly worth attending.
In addition, the Beamer slides made by the students would also be made available within a week of the lecture.

Tutorials/Assignments/Projects

The final “project” was in fact a paper review of a contemporary development in game theory followed by a presentation. The professor allowed considerable freedom in choice of topics, so there were topics as varied as deep reinforcement learning, financial market models, poker theory, spacecraft rendezvous and so on. Students had at least a month to read the paper and summarize it in a few slides.

Feedback on Exams

The only exams were the take-home quizzes. For the most part, the questions were quite easy and based on the class discussions. There was no time pressure either, so one could give the exam based on convenience.

Motivation for taking this course

I had already taken SC607 (Optimization) under Prof. Ankur Kulkarni the previous semester, so I felt that this was a logical step in that direction. I was also interested in exploring topics in the domain of applied data science and finance, so game theory came as a natural choice. An added benefit was the freedom in tagging it as a minor, DE or ALC.

Course Highlights

The last part of the course on dynamic multi-agent games is quite interesting and has applications in many fields.

Course Importance

For starters, it is a part of the DS minor basket. Unsurprisingly, it has a lot of relevance in reinforcement learning and other multi-agent as well as adversarial learning problems.
Although the course leans towards the informational side of game theory, the general ideas are also very handy in finance and economics.

How strongly would I recommend this course?

Quite strongly, game theory is everywhere. If not for the various applications mentioned several times in this review, it helps you think more analytically while trying to solve simple day-to-day problems :-p

When to take this course?

5th semester. Ideal semester would be 5th or 7th.

Going Forward

For those interested in the AI/ML domain, it may be a good idea to pick up CS747 parallelly, which can then be followed by CS748 in the autumn semester.

References Used

None as such, lecture notes are sufficient.

Other Remarks

Some of the equivalent courses in this domain are IE611 (Decision Analysis and Game Theory) and HS402 (Game Theory and Economic Analysis).

SC 631 Review By: Aditya Iyengar